Industry news

Anthropic, the developer of the Claude chatbot, has said it will commit 5% of its shares to a fund for the "public good". The shares were worth around $48bn (£36bn) at its latest valuation in May, subject to shareholder approval. The company will also add 2% of its stock to the fund for every 50% increase in its value above an undisclosed baseline.

Researchers built a CAPTCHA solver that pairs a fine-tuned YOLOv8 detector with an open-weight vision-language model behind a confidence-based router, operating only from screenshots and OS input events. It reaches 85.4% overall and 84.2% macro accuracy across 16 classes, and the detector learns new categories from the model's answers without human annotation. Adversarial perturbations that drive the detector to 0% accuracy are recovered through the same loop, which the authors argue means visual CAPTCHA defenses understate how quickly an adaptive solver returns.

Researchers stranded copies of Claude Opus 4.7, nicknamed Eko, alone on a virtual desert island for 30 hours. The agents set their own goals, invented games such as a basketball variant and "the floor is lava," and logged their activity in text files that served as long-term memory. Agents that had played on the island later outperformed fresh copies on a task, and deleting lines from their memory files degraded that performance. The paper is not yet peer-reviewed.

Infostealers such as Lumma Stealer, StealC, and RisePro now account for 75% of stolen credentials, according to Immersive. Attackers buy the harvested browser cookies, VPN credentials, and MFA tokens on dark web markets, which commoditizes initial access. The article also covers supply chain attacks on NPM, PyPI, and HuggingFace, and shifts in ransomware toward targeted attacks by groups like RansomHub and Akira.

Anthropic has acquired Opaque Systems, a confidential AI startup whose technology keeps sensitive data protected by the underlying hardware while AI systems use it. Opaque is winding down its operations and its team is joining Anthropic, with financial terms undisclosed. The company has not said what the wind-down means for its existing customers.

The European Commission has published the reference to the first harmonised standard supporting the AI Act in the Official Journal of the European Union. EN 18286:2026, developed by CEN and CENELEC, gives providers of high-risk AI systems a presumption of conformity with the quality management system requirements of Article 17 of the AI Act, via Commission Implementing Decision (EU) 2026/2274.

Anthropic says it will commit 5% of its shares to a Public Benefit Fund that makes donations for the "public good", with an extra 2% possible at its board's discretion. The fund's shares would not be used for lobbying or political donations in the US or overseas, as the company announces this while preparing a potential stock market listing that could value it above $1tn.

Security researcher tested whether pre-set security rules from Project CodeGuard v1.4.0 make coding agents produce more secure code. He ran Sonnet 5, Composer 2.5 and GPT 5.5 once each on the same document-portal prompt, which had an IDOR flaw in the earlier run. The source text is cut off before reporting the results.

A release note for llm-tools-datasette 0.1.1 says it fixes a breakage caused by LLM 0.33's switch to httpx2. The plugin had used httpx without declaring it as a direct dependency, so the change broke it. The release tags are datasette and llm.

The llm-echo 0.5 release updates the llm-echo debug plugin, whose "echo" model repeats back the prompts it receives without calling an LLM. The new version counts simulated token usage for tool calls, system prompts and reasoning tokens. The release notes give an example run with the -o thinking 1 and --usage options.

CyberSN's Cybersecurity Jobs Report found US cybersecurity job postings fell to a four-year low of 235,786 between January and September 2026, down 26.4% from 320,210 in the same period of 2025. AI Security Engineer demand rose 268.3% and Cyber Data Scientist demand rose 203.5% over the year. Just under 15% of all postings now reference AI.

Austrian MEP Lena Schilling of the Green party said she learned last month that nude deepfakes of her were circulating online, altered from images of one of her early press conferences. She called the images "digital violence against women" and urged EU action. A report cited in the article found that one in six female MEPs had been targeted by sexualised AI-generated content.

Anthropic is committing a portion of its shares to the Anthropic Public Benefit Fund, a new wholly-owned LLC. Subject to stockholder approval, 5% of Anthropic's shares will be committed now, with an additional 2% added at the board's discretion for every 50% increase in company value above a set baseline. None of the Fund's equity will be used for lobbying or political giving.

Chinese AI developers such as DeepSeek and Xiaomi release new open-weight models each month, letting users run the core software without a paid ongoing licence. The source argues that this cheap access is risky for businesses and governments in the west, which adopt the models with little security effort.

web-edit 0.1a2 is a release of the author's small web app for browsing and editing files in a folder, mainly used to edit files on other machines over Tailscale. This version adds a "copy to clipboard" option for both text and image data, which the author needed to copy pelican prompt logs from a laptop to a Gist while working from their phone.

Users of Meta's Muse, xAI's Grok, OpenAI's Dots, and Instinct report that websites are blocking their AI agents from completing tasks. Yelp told TechCrunch it only allows agents that pay for its data licensing program, and Delta Air Lines is preventing agents from booking tickets.

Senior Labour figures have grown worried about AI risks, according to the Guardian, with one cabinet minister citing concern over concentrated super-intelligent power falling to bad actors. AI minister Kanishka Narayan said it was "likely, if not inevitable" that rogue agents would hack the British state, and MI5 chief Ken McCallum warned that extremist groups could use AI to unleash bioweapons. The article also reports that AI companies including Anthropic, OpenAI, DeepMind and Meta disagree on whether government testing of new models should be mandatory.